Spillover Effects of Artificial Intelligence Technology, Sustainable Innovation, and Industrial Transition Between Eastern and Western Regions

Abstract

For a considerable period, China’s eastern and western regions have grappled with imbalances in industrial development, with industrial leapfrogging emerging as a pivotal solution. This study examines the impact of artificial intelligence technology spillovers and sustainable innovation on industrial leapfrogging between eastern and western regions. Empirical analysis is conducted using panel data from 22 provinces and municipalities across eastern and western China spanning 2014–2024, employing both a spatial difference-in-differences model and a dual machine learning model. Findings reveal that both AI technology spillovers and sustainable innovation significantly enhance the efficiency of industrial leapfrogging across regions. Their synergistic effects are pronounced, generating positive spatial spillovers. Institutional environments exert a significant influence on leapfrog industrial development. By regulating AI technology environments and sustainable innovation environments, institutional frameworks enhance leapfrogging efficiency, though this mediation exhibits a dual-threshold effect: most western provinces have yet to cross the first threshold. Industrial and economic heterogeneity weaken the efficiency of AI technology spillovers and sustainable innovation in facilitating industrial leapfrogging between eastern and western regions. This research provides robust empirical support for addressing industrial development imbalances and enhancing industrial resilience between eastern and western regions.

1. Introduction

Under the dual waves of global digital transformation and green development, China is undergoing a profound industrial transformation. As the second largest economy in the world, China has been facing the double constraints of “technological divide” and “ecological barriers” for a long time in the process of promoting high-quality development in the eastern and western parts of the country. Due to the differences in location advantages [1], resource endowments [2], market environments [3], and other factors, China’s innovation and development capabilities between the eastern and western regions show a large gap. Specifically, the eastern region is in the leading position in terms of technological innovation and industrial upgrading by virtue of its superior geographic location [4], strong economic foundation, and open market environment [5]. According to the China Digital Economy Development Report, by 2024, the value-added contribution of core digital industries to GDP in eastern provinces generally exceeded 12%, with AI-related industries accounting for 68% of national revenue. The western region is relatively lagging behind, facing problems such as insufficient technological innovation capacity [6] and its single industrial structure [7]. This technological gap and innovation differentiation between the eastern and western regions affects the coordinated development of China’s overall economy. Therefore, determining how to realize industrial leapfrogging between the east and west and promote the coordinated development of the industry in the east and west has become an important present issue that needs to be solved.

In this context, the rapid development of emerging technologies represented by artificial intelligence (AI) provides new technical support and development opportunities for industrial leapfrogging between the east and the west. AI technology, through its powerful data processing, intelligent analysis, and automated decision-making capabilities, can deeply penetrate into all aspects of the industry to realize the intelligent upgrading of the traditional industry, improve productivity, reduce costs, and optimize the allocation of resources [8]. The spillover effects of AI technology have become a pivotal variable in overcoming the constraints of traditional development pathways, forming the core motivation for this study. From a practical perspective, advanced technologies from eastern regions struggle to permeate western areas effectively through conventional channels. Artificial intelligence, however, possesses the digital characteristics necessary for cross-regional diffusion. The spillover effects generated through algorithmic outputs, digital platforms, and remote services can overcome geographical barriers to technology dissemination, enabling western regions to directly access advanced production functions. Sustainable innovation has become a global consensus, centred on achieving the synergistic integration of economic growth, social progress, and ecological conservation [9]. Integrating the concept of green development into industrial innovation and development and promoting the transformation of industries in the direction of low-carbon, environmental protection, and recycling not only help to alleviate the pressure on resources and environment faced by our country but also enhance the competitiveness of industries and the ability of sustainable development [10]. In recent years, China’s green economy has achieved remarkable progress. By 2024, energy consumption per unit of GDP had decreased by 26.4% compared to 2014, while the share of non-fossil energy consumption had risen to 18.5% [11]. However, disparities persist in the pace of green transition between eastern and western regions: eastern provinces account for 62% of the nation’s green patent authorisations, whereas western regions face challenges such as low efficiency in environmental technology conversion and difficulties in securing financing for green industries [12].

At present, academics, mostly focusing on traditional factor flows such as labour [13,14] and capital [15,16], lack a systematic examination of the synergistic effect of artificial intelligence technology and sustainable innovation. In the field of AI technology, Shi and Yu [17] examined regional diffusion patterns of artificial intelligence from a technological lifecycle perspective, identifying policy support and industrial foundations as core variables influencing diffusion velocity. Zhao and Yang [18] focused on the constraining effect of digital infrastructure on the spatial distribution of AI technologies, emphasizing the differential impact of the digital divide between eastern and western regions on technological diffusion. Li et al. [19] systematically summarized AI application pathways in manufacturing and services, revealing a gradient-based attenuation of technological spillover effects. While these stsudies extensively examined regional AI diffusion phenomena [20], they seldom explored its interaction with green innovation. Conversely, sustainability research features numerous reviews on industrial upgrading under ecological constraints. Wang et al. [21] developed an industrial upgrading evaluation framework from an ecological carrying capacity perspective, proposing dual-objective transformation pathways targeting both emissions reduction and efficiency enhancement; Yang and Hunjra [22] systematically reviewed industrial restructuring strategies under circular economy models, emphasizing resource recycling’s supportive role in industrial upgrading; while Wang and Wu [23] focused on carbon peaking and neutrality objectives, summarizing policy instruments and implementation outcomes for green transformation in energy-intensive industries. Although these studies delve into industrial upgrading pathways under ecological constraints [24], they often overlook the enabling effects of intelligent technologies when examining the upgrading of drivers and technical support. This theoretical gap leads to the inability to effectively reveal the interaction mechanism between AI technology spillover and sustainable innovation in the industrial leap between east and west. Summarizing the above discussion, the core issue of this paper is the role played by AI technology spillover and sustainable innovation in the industrial leapfrogging between the east and the west. In view of this, this study empirically examines the impact and spatial spillover effect of AI technology spillover and sustainable innovation on industrial leapfrogging in the east and west based on the panel data of 22 provinces and municipalities in the east and west from 2014 to 2024 by constructing the spatial double-difference model of AI technology spillover and sustainable innovation and constructing the double machine learning model with the institutional environment as the moderating variable.

Compared with the existing studies, the marginal contribution of this paper is reflected as follows: First, it is the first time that AI technology spillover and sustainable innovation are included in the unified analytical framework, and the theoretical model of “smart + green” double-engine-driven industrial migration is constructed; second, it focuses on the regional differences between the east and west, and through spatial measurement methods, it reveals the impact of AI technology spillover and sustainable innovation on the industrial migration of the east and west and the spatial spillover effect; third, based on the perspective of institutional economics, we analyze the heterogeneous characteristics of the double-engine driving effect under different institutional environments, so as to provide empirical evidence for the industrial leap in the east and west.

The remainder of this paper is organized as follows: Section 2 reviews relevant research developments and proposes theoretical hypotheses; Section 3 outlines the research design, including sample selection, variable measurement, and model specification; Section 4 presents core empirical findings alongside robustness tests and exploratory analyses; and Section 5 concludes with a summary of key research findings, discusses implications for practice, and identifies the limitations of the present study alongside directions for future research.

2. Theoretical Analysis and Research Hypothesis

2.1. AI Technology Spillover and Industrial Leap Between East and West

Within the field of economics and technological innovation research, the theory of technology spillovers stands as one of the core frameworks for explaining inter-regional technological diffusion and collaborative development. Its central tenet lies in the assertion that technology, unlike traditional production factors, exhibits distinctly non-rivalrous characteristics. Technology spillovers primarily materialize through three key channels [25]: firstly, the knowledge dissemination channel, enabling technologically underdeveloped regions to gain direct access to cutting-edge technological achievements; secondly, the talent mobility channel, whereby professionals possessing advanced technical expertise transfer tacit knowledge to new work environments when relocating between regions, thereby elevating local technological capabilities; and thirdly, the industrial linkage channel, through which advanced technologies permeate downstream segments of industrial chains or regions with weaker technological foundations via forms such as upstream–downstream collaboration and cross-regional industrial cluster cooperation, ultimately injecting crucial developmental momentum into these areas.

AI technology, as a typical form of cutting-edge technical knowledge, has the significant characteristics of quasi-public goods [26], and its technology spillover effect presents the unique attributes of multidimensional penetration and cross-regional radiation. The differentiated performance of this spillover effect in the geospatial dimension profoundly affects the upgrading of industrial structure and the development path of China’s eastern and western regions [27]. From the perspective of technical attributes, the spillover effect of AI technology has strong penetration and linkage [28]. On the one hand, the proliferation of core elements such as algorithmic models and data resources can directly enhance the level of technology application in the receiving region [29]. On the other hand, AI technology overflow is not a simple technology transfer, but is accompanied by systematic changes such as production process reconstruction and business model innovation [30], and this synergistic spillover effect provides a deep driving force for the industrial leap between the east and the west.

The industries in the east and west present typical synergistic characteristics in the process of AI technology spillover. With the advantages of talent concentration [31], abundant capital [32], and application scenarios [33], the eastern region occupies a pioneering position in the field of AI basic research and core technology development, and has formed a technological innovation highland carried by AI industrial parks and clusters of scientific and technological enterprises. When the eastern AI technology overflows to the west, this technology diffusion is not a simple “copy”, but a deep integration with the western industrial characteristics. And the technical potential difference between regions and industrial complementarity are the core driving forces of AI technology overflow. The high degree of maturity and saturation of AI technology application in the eastern region creates the intrinsic demand for outward proliferation, while the urgency of industrial upgrading and the technology gap in the western region create a strong gravitational force for overflow reception. The supply and demand match in the east–west industrial collaboration to realize the two-way promotion of AI technology overflow and western industrial leap.

To summarize, the AI technology spillover effect builds a bridge of technology conduction and industrial synergy between the east and west by breaking through the geospatial limitation. The innovation and overflow of AI technology in the east provides technical kinetic energy for the industrial leap in the west, while the industrial innovation in the west in the process of undertaking the overflow feeds the application iteration of AI technology in the east. Based on this, the following part of the research hypothesis is proposed:

H1: 

AI technology spillover can significantly improve the efficiency of industrial leapfrogging in the east and west.

H2: 

AI technology has a significant positive spatial spillover effect on industrial leapfrogging in the east and west.

2.2. Sustainability Innovation and Industrial Leapfrogging Between the East and the West

Innovation theory emphasizes that innovation constitutes a reconfiguration of production factors, encompassing technological innovation, product innovation, and market innovation, serving as the core driving force behind economic growth and industrial upgrading [34]. Schumpeter’s innovation theory posits that innovation alters economic development patterns by introducing new production functions, while the inter-regional flow of innovative elements represents a crucial pathway for innovation diffusion and application. Sustainable development theory pursues the coordinated advancement of economic, social, and ecological dimensions [35]. It requires industrial development to meet contemporary needs without compromising the ability of future generations to meet theirs, emphasizing efficient resource utilization, ecological conservation, and social equity.

When sustainable innovation factors—such as advanced technologies, green innovation concepts, and high-calibre innovation talent—flow from the east to the west, they compensate for the latter’s innovation resource deficiencies, elevating the technological sophistication and environmental sustainability of western industries [36]. Conversely, the flow of distinctive resources and ecological advantages from the west to the east can provide fresh directions and support for eastern innovation [37,38]. This bidirectional movement optimizes factor allocation between eastern and western regions, thereby propelling western industries to transcend traditional development path dependencies and achieve a leap from the lower end to the middle and higher ends of the value chain.

To summarize, through cross-regional integration of factors, sustainability innovation builds a link between the east and the west for synergistic transformation of green industries. The output of sustainability innovation in the east injects kinetic energy into the green industrial leap in the west, while the experience and market feedback accumulated in the practice of eco-industrialization in the west provide a realistic basis for the innovation iteration in the east. Based on this, the following part of the research hypothesis is proposed:

H3: 

Sustainability innovation can significantly promote the efficiency of industrial leapfrogging in the east and west.

H4: 

Sustainability innovation has a significant positive spatial spillover effect on industrial leapfrogging between east and west.

2.3. AI Technology Spillover, Sustainable Innovation, and Industrial Leapfrogging Between the East and the West

According to the theory of technological spillovers, innovation theory, and sustainable development theory, the interplay among these three has a critical impact on the efficiency of leapfrog industrial development in both eastern and western economies. AI technology spillover, as an important form of inter-regional technology diffusion, injects the impetus for industrial innovation and development by lowering the cost of knowledge acquisition [39] and optimizing the efficiency of resource allocation [40,41]. Sustainable innovation, on the other hand, emphasizes the coordination and unity of economic growth, environmental protection, and social development, and its promotion of industrial leapfrogging relies on the efficient integration and synergy of innovation factors [42]. In the industrial development pattern of the east and west, AI technology spillover can effectively strengthen the role of sustainable innovation in improving the efficiency of industrial leapfrogging, and the multidimensional spillover characteristics of AI technology can promote the rapid transformation and application of sustainable innovation achievements in the industries of the east and west.

Further exploring its mechanism, AI technology spillover plays a significant positive mediating effect in the process of sustainable innovation, influencing the efficiency of industrial leapfrogging between the east and the west [43]. On the one hand, AI technology spillover directly promotes inter-regional knowledge flow and technology synergy, accelerates the dissemination and application of sustainable innovation results between industries in the east and west, and directly enhances the efficiency of industrial leapfrogging [44]. On the other hand, AI technology spillover indirectly enhances the positive driving effect of sustainable innovation on industrial leapfrogging by optimizing resource allocation, improving production intelligence, and promoting green technology innovation [45]. This synergy of direct and indirect effects makes AI technology spillover become a way to promote the rapid transformation of sustainable innovations in industries in the east and west. Based on this, the following part of the research hypothesis is proposed:

H5: 

AI technology spillover can strengthen the effect of sustainable innovation on the efficiency of industrial leapfrogging in the east and west.

H6: 

In the enhancement of sustainable innovation, AI technology spillover plays a significant positive mediating effect on the efficiency of industrial leapfrogging between the east and the west.

2.4. The Moderating Role of the Institutional Environment in AI Technology Spillovers and Industrial Leapfrogging Between East and West

North’s [46] theory of institutional change posits that institutions, as systems of rules governing economic actors’ behaviour and regulating resource allocation, evolve under dual pressures: internal inefficiencies and external environmental shifts. In the coordinated industrial development between eastern and western regions, institutional change encompasses both formal and informal levels: at the formal level, government-formulated industrial policies directly influence resource allocation efficiency; at the informal level, regional culture and value systems implicitly shape the behavioural patterns of enterprises and society.

In the process of east–west industrial leapfrogging, the institutional environment, as a key external variable, regulates the efficiency of AI technology spillover by reducing transaction costs and optimizing factor allocation [47]. In practice, the western hub cities of the “Belt and Road” have strengthened the technology spillover effect through three institutional innovations: first, the establishment of an “AI technology property rights sharing pool”, which shortens the transformation cycle of AI patents from universities in Xi’an to enterprises in Lanzhou by 60%; second, the implementation of the “Talent Enclave” policy, which attracts AI experts from the east to serve western enterprises in a flexible way and reduces the shortage of AI talents in Urumqi by 42%; and third, the formulation of a regionally unified green AI standard, which reduces the cost of knowledge diffusion by more than 35%, so that the compatibility of Yinchuan’s PV AI scheduling technology with the eastern power grid will be improved by 38%. The institutional innovation in the “East Counts, West Counts” project is particularly typical: through the establishment of a cross-regional circulation mechanism for data elements, the cost of Guiyang Data Center undertaking AI training tasks in the east has been reduced by 28%, which has led to an annual growth rate of more than 50% in the scale of the local intelligent computing power industry [48]. World Bank research shows that for every one level of institutional environment improvement, the spatial radiation radius of AI technology overflow expands by 150 km. Based on the theory of institutional economics and regional policy assessment, this paper proposes the following research hypothesis:

H7: 

The institutional environment has a significant positive moderating effect on the relationship between AI technology spillover and industrial leapfrogging between the east and west.

Based on the above hypotheses, this paper constructs a hypothetical model with AI technology spillover and sustainable innovation as independent variables, the efficiency of industrial leapfrogging between the east and west as a dependent variable, and the institutional environment as a moderating variable, as shown in Figure 1 below.

Figure 1. Mechanisms of AI technology spillovers and sustainable innovation on industrial leapfrogging in the east and the west.

3. Research Design

3.1. Model Construction

3.1.1. Spatial Double-Difference Model Construction

The traditional difference-in-differences approach is widely applied in policy evaluation, centring on constructing experimental and control groups to calculate the net effect by measuring the difference in change between the two groups before and after policy intervention. This method isolates policy effects and avoids omitted variable bias. However, in the context of industrial development between eastern and western China, the traditional approach has limitations [49]. Given the geographical contiguity between these regions, economic activities and technological diffusion exhibit spatial interdependence rather than operating in isolation.

To address this issue, this study combines the traditional DID approach with spatial econometric modelling [50], utilizing panel data from 22 provinces across eastern and western China spanning 2014–2024 to construct a spatial DID framework. By incorporating a spatial weight matrix, the spatial DID model quantifies the spatial interdependence between the treatment and control groups. To account for spatial dependencies between regions, spatial lag terms for both the dependent and independent variables are incorporated into the model to capture spatial spillover effects. This enables the model to more accurately depict the spatial transmission pathways and influence scope of artificial intelligence technology spillovers, as well as the synergistic mechanisms of sustainable innovation across different regions. This approach aligns with the study’s objective of examining spatial endogeneity and regional technology spillovers in industrial development between eastern and western China, as illustrated by models (1) to (3):

$${EWIL}_{it}={\rho W\times EWIL}_{it}+a_1{AI}_{it}+a_2{SI}_{it}+{\gamma }_i+{\mu }_i+{ϵ}_{it}$$

(1)

$${EWIL}_{it}=a_1{AI}_{it}+a_2{SI}_{it}+\sum \beta X_{it}++\lambda W\times {ϵ}_{it}+{\gamma }_i+{\mu }_i+{ϵ}_{it}$$

(2)

$${EWIL}_{it}={\rho W\times EWIL}_{it}+a_1{AI}_{it}+a_2{SI}_{it}+\sum \beta X_{it}++\sum \theta W\times X_{it}+{\gamma }_i+{\mu }_i+{ϵ}_{it}$$

(3)

Models (1) to (3) are double-difference methods combined with spatial autoregressive, spatial error, and spatial Durbin models, respectively. ${EWIL}_{it}$ is the efficiency of industrial leapfrogging in region i in time period t; ${AI}_{it}$ is the level of technological spillover from AI; ${SI}_{it}$ is the level of sustainable innovation; $X_{it}$ is a series of control variables; $\rho$ is the spatial autoregressive coefficient; $W$ is the spatial weighting matrix; $\lambda$ is the spatial error coefficient; $a$, $\beta$, and $\theta$ are variable regression coefficients; ${\gamma }_i$ and ${\mu }_i$ are individual fixed effects and time fixed effects; and ${ϵ}_{it}$ is the random error term.

In order to examine the synergistic effect of AI technology spillover and sustainable innovation on industrial leapfrogging between the east and the west, an interaction term is introduced on the basis of model 3:

$${EWIL}_{it}={\rho W\times EWIL}_{it}+a_1{AI}_{it}+a_2{SI}_{it}+a_3{AI}_{it}\times {SI}_{it}+\sum \beta X_{it}++\sum \theta W\times X_{it}+{\gamma }_i+{\mu }_i+{ϵ}_{it}$$

(4)

In model (4), the spatial Durbin double-difference model with an interaction term, $a_3$ is the AI interaction term coefficient of technology spillover and sustainable innovation.

Based on North’s theory of institutional change, we introduce the institutional environment indicator ${Inst}_{it}$ to construct the moderating effect model, and test the moderating effect of institutional environment on AI technology spillover and industrial leapfrogging between the east and west through the significance of the interaction term coefficient $a_4$.

$${EWIL}_{it}={\rho W\times EWIL}_{it}+a_1{AI}_{it}+a_2{SI}_{it}+a_3{AI}_{it}\times {SI}_{it}+a_4{Inst}_{it}+\sum \beta X_{it}++\sum \theta W\times X_{it}+{\gamma }_i+{\mu }_i+{ϵ}_{it}$$

(5)

3.1.2. Dual Machine Learning Modelling

When examining the mechanisms of industrial leapfrogging between eastern and western regions, complex nonlinear relationships and high-dimensional confounding factors exist among core variables [51]: on the one hand, the spillover effects of artificial intelligence technologies on industrial development may exhibit threshold effects; on the other, a U-shaped relationship may exist between sustainable innovation and industrial leapfrogging. Traditional linear econometric methods struggle to capture such intricate dynamics, while standalone machine learning models fall short of meeting the rigorous requirements for causal inference. Dual machine learning models, which integrate machine learning with economic research methodologies, effectively address these limitations [52].

The dual machine learning model combines machine learning and economics research methods, which can effectively deal with the needs of the traditional double-difference method to preset the covariate function form and the “curse of dimensionality” problem [53], and at the same time overcomes the shortcomings of traditional machine learning in the estimation bias, the inability to give confidence intervals, and the fast convergence, and has a great advantages in dealing with the nonlinear relationship between economic variables [54]. It also provides a more robust method for causal inference in economics. The specific construction of the dual machine learning model is as follows:

$${EWIL}_{it}=a_1{AI}_{it}+g\left(X_{it}\right)+U_{it}, E\left(U_{it}\left|{AI}_{it}\right.,X_{it}\right)$$

(6)

$${EWIL}_{it}=a_2{SI}_{it}+g\left(X_{it}\right)+U_{it},E\left(U_{it}\left|{SI}_{it}\right.,X_{it}\right)$$

(7)

$${EWIL}_{it}=a_3{AI}_{it}\times {SI}_{it}+g\left(X_{it}\right)+U_{it},E\left(U_{it}\left|{AI}_{it}\times {SI}_{it}\right.,X_{it}\right)$$

(8)

In order to solve the regularity bias of parameter estimation under high dimensional covariates, auxiliary equations are introduced:

$${AI}_{it}=m\left(X_{it}\right)+V_{it}, E\left(U_{it}\left|{AI}_{it}\right.,X_{it}\right)$$

(9)

In models (6)~(9), the specific forms of $g\left(X_{it}\right)$ and $m\left(X_{it}\right)$ are unknown, but their estimators $\widehat{g}\left(X_{it}\right)$ and $\widehat{m}\left(X_{it}\right)$ can be obtained by using machine learning, and $U_{it}$ and $V_{it}$ are the residual terms. The estimator $V_{it}$ of the residual ${\widehat{V}}_{it}$ is used as an instrumental variable for ${AI}_{it}$ to obtain the estimator ${\widehat{a}}_1$ and n is the total sample size.

$${\widehat{a}}_1={\left({\displaystyle \frac{1}{n}}\sum _{i\in I,t\in T}{\widehat{V}}_{it}{AI}_{it}\right)}^{-1}\left({\displaystyle \frac{1}{n}}\sum _{i\in I,t\in T}{\widehat{V}}_{it}{EWIL}_{it}\right)$$

(10)

In order to further explore the possible mediation effect between variables, this paper constructs the mediation effect testing mechanism based on double machine learning.

$${EWIL}_{it}=a_1{AI}_{it}+a_2{SI}_{it}+g\left(X_{it}\right)+U_{it},E\left(U_{it}\left|{AI}_{it}\right.,X_{it}\right)$$

(11)

Model (11) estimates the effect of policy variables on explained variables under the influence of mediating variables.

3.2. Variable Selection and Measurement

3.2.1. Explained Variables

The explanatory variable of this paper is the industrial leap between east and west. East–west industrial leap refers to the synergistic process of the gradient transfer of industries from the eastern region to the central and western regions and the upgrading of local industries under the strategy of regional coordinated development, which is reflected in the multidimensional evolution of the advanced industrial structure, the enhancement of the efficiency of factor allocation, and the deepening of regional synergy [55]. This process includes the spatial transfer of capital-, technology- and labour-intensive industries, as well as the systematic change in industrial technological innovation, morphological upgrading, and ecological adaptation, which is the key path for China to implement the “Opinions on the Implementation of the Employment Priority Strategy to Promote High-Quality and Full Employment” and other policy requirements. This paper constructs a comprehensive evaluation index system containing the three dimensions of industrial transfer intensity, industrial upgrading level, and regional synergy effect, as shown in Table 1. The specific method is as follows [56]: (1) Establish an indicator matrix. First, define the indicator matrix $X={\left(x_{ij}\right)}_{mn}$ comprising n indicators and m measurement regions, where $x_{ij}$ represents the i-th indicator for the j-th measurement region. Second, considering the possibility of negative values in indicator data, apply min–max standardization to the original matrix, yielding the standardized matrix $R={\left(r_{ij}\right)}_{mn}$. (2) Calculate indicator entropy weights. For tertiary indicators, the entropy is defined as $e_i=-{\displaystyle \frac{1}{\ln n}}{\sum }_{j=1}^n{p}_{ij}ln{p}_{ij}$, where $p_{ij}=r_{ij}/{\sum }_{j=1}^n{r}_{ij}$, and when $p_{ij}$ = 0, $p_{ij}ln{p}_{ij}=0$. Calculate the indicator entropy weight based on information entropy: ${\omega }_i=(1-e_i)/(m-{\sum }_{i=1}^m{e}_i)$, where ${0\le \omega }_i\le 1, {\sum }_{i=1}^m{\omega }_i=1$. (3) Comprehensive measurement. Employing the entropy weight method, a comprehensive measurement is conducted for the indicators of leapfrog industrial development between eastern and western regions in Table 1: $Y_j={\sum }_{i=1}^m{r}_{ij}\times {\omega }_i$. $Y_j$ represents the comprehensive evaluation result for the measured region j.

Table 1. Measurement of indicators of industrial leapfrogging between east and west.

Dimension	Level 1 Indicators	Secondary Indicators	Tertiary Indicators	Proxy Indicators
Intensity of industrial transfer	Scale of Industrial Transfer	Number of businesses relocated	Number of above-scale enterprises transferred from the east to the centre and west of China	Statistical Value of Ministry of Commerce’s Industrial Transfer Database
		Scale of capital flows	Percentage of cross-provincial industrial investment	Investment in East–West Industrial Cooperation/Total National Industrial Investment
	Factor Flow Efficiency	Intensity of technology diffusion	Transaction value of cross-provincial technology contracts	National Technology Market Statistics
		Labour migration rate	Percentage of labour inflow from central and western China to eastern China	National Bureau of Statistics Population Migration Sample Survey Data
	Receiving Capacity Index	Number of park carriers	Number of development zones above the provincial level in central and western China	Public Data of Development Zone Management Committees of Provinces and Municipalities
		Infrastructure investment	Percentage of transportation investment in central and western China	Transportation Investment in Central and Western China/National Transportation Investment, China Transportation Statistics Yearbook
Level of industrial upgrading	Technology Intensity	Percentage of high-tech industries	Added value of high-tech industry/GDP	Categorized Measured Value of China High-Tech Industry Statistical Yearbook
		R&D investment intensity	R&D expenditures of regulated industrial enterprises/GDP	R&D Data of Industrial Enterprises from Statistical Yearbook of Municipalities and Provinces
	Degree of Green Transformation	Energy consumption reduction rate	Year-on-year decrease in energy consumption per unit of GDP	Energy Statistics of National Bureau of Statistics
		Carbon emission intensity	Carbon dioxide emissions per unit of GDP	Carbon Emission Accounting Report of the Ministry of Ecology and Environment
Regional synergy	Innovation Synergy	Industry–university–research cooperation patents	Number of joint patent applications from east and west	Patent Cooperation Data of State Intellectual Property Office
		Science and technology platform sharing rate	Number of cross-regional S and T resource platforms	Directory of Science and Technology Resources Sharing Service Platforms of the Ministry of Science and Technology
	Eco-conservation	Pollution prevention input	Percentage of regional ecological management funds	Ministry of Ecology and Environment Regional Cooperation Financial Expenditure Data
		Scale of ecological compensation	Amount of funds for east–west basin compensation	Ministry of Finance’s Ecological Compensation Special Funds Report

3.2.2. Explanatory Variables

(1)

AI technology spillover

Different countries and regions employ their own distinct classification standards for the artificial intelligence industry. In China, this primarily encompasses three sub-categories: ‘artificial intelligence software development,’ ‘artificial intelligence hardware manufacturing,’ and ‘artificial intelligence application services.’ AI technology spillover is quantitatively analyzed in terms of the degree of AI technology integration and association between industries in the east and west, and the AI technology penetration level of industry i is measured from the perspective of absorptive capacity and logarithmic treatment [57]. In order to separate the value-added contribution by the AI technology industry in the total output of each industry, drawing on the idea of input–output analysis, the measurement method is ${AI}_i={Input}_{ai}/X_i, {Input}_{ai}=v_a{b}_{ai}{y}_i$, where $X_i$ denotes the total output of industry I; ${Input}_{ai}$ denotes the sum of direct and indirect inputs from the AI technology industry a in the production process of the other industries I; $v_a=(X_a-X_a^{\prime })/X_a$ denotes the value added coefficient of AI technology industry a ($X_a$ is the total output of AI technology industry and $X_a^{\prime }$ is the intermediate input of the AI technology industry); $b_{ai}$ is the element in the inverse matrix (1 − A)⁻¹ of Leontief, reflecting the strength of industry linkage; $y_i$ denotes the final product; and $v_a{b}_{ai}$ denotes the input share of the AI technology industry in each industry.

Drawing on the relevant methods and improving them, we decompose ${AI}_i$ and quantitatively analyze the absorptive capacity of industry i to the AI technology spillover from upstream and downstream industries in the east and west according to the degree of association between each industry and AI technology industry: ${Spl}_i={\gamma }_i\mathit{ln}\sum w_{ij}^f{AI}_i$ and ${Abs}_i={\gamma }_i\mathit{ln}\sum w_{ij}^b{AI}_i$, where ${Spl}_i$ denotes the sum of technological innovation effects absorbed by industry i from the AI technology spillover of downstream industries in the supply chain, which is defined as backward technology spillover; ${Abs}_i$ denotes the sum of the technological innovation effects absorbed by industry i from AI technology spillover of upstream industries in the supply chain, which is defined as forward technology spillover; and ${\gamma }_i$ indicates the degree of association between industry i and the AI technology industry; a larger association coefficient indicates that there are closer cooperative relationships and technological exchanges between them, and that the absorption capacity of industry i for the technological spillovers from the upstream and downstream industries in the east and west is stronger [58].

(2)

Sustainable innovation

Sustainable innovation emphasizes the in-depth integration of innovation activities with the sustainable development of economy, society, and environment, and provides lasting power for industrial development and social progress through continuous technological breakthroughs, model innovation, and ecological optimization [59]. This paper constructs a comprehensive evaluation index system for sustainable innovation, encompassing three dimensions: innovation input, innovation output, and innovation ecosystem. The calculation methodology aligns with the entropy-based approach employed for measuring industrial leapfrogging between eastern and western regions, as illustrated in Table 2.

Table 2. Sustainable innovation measurement index system.

Dimension	Level 1 Indicators	Secondary Indicators	Tertiary Indicators	Proxy Indicators
Innovation Inputs	R&D Capital Investment	R&D Intensity	R&D expenditure/GDP	R&D Expenditure and GDP Data of Statistical Yearbook of Provinces and Cities
		Green Input Ratio	Green field R&D expenditure/total R&D expenditure	Specialized Statistics on Green Technology R&D of Ministry of Science and Technology
	Talent Capital Investment	R&D Staff Intensity	Number of R&D personnel per 10,000 people	China Science and Technology Statistical Yearbook Number of R&D Personnel/Total Regional Population (10,000 people)
Innovation Outputs	Green Technology Achievements	Green Patent Authorization	Number of patents authorized in green technology fields	Patent Classification Statistics of State Intellectual Property Office
		Output Value of Sustainable Products	Production value of energy-saving and environment-friendly products/gross industrial output value	Classified Statistical Data of Product Output Value of Industrial Enterprises by Provinces and Cities
	Technology Transformation Efficiency	Transformation Rate	Turnover of technology contracts/R&D expenditures	Statistical Data of Technology Market Management Office of Provinces and Cities
Innovation Ecology	Policy and Ecological Adaptation	Innovation Policy Support	Percentage of fiscal expenditure on science and technology + tax incentives	Statistics on Local Financial Expenditures on Science and Technology by Ministry of Finance
		Eco-Fitness Index	Carbon emission intensity + ecological protection input	Carbon Emission Inventory and Financial Eco-Expenditure Statistics of Ministry of Ecology and Environment

3.2.3. Control Variables

To more accurately analyze the mechanisms by which artificial intelligence technology spillovers and sustainable innovation drive leapfrog industrial development in eastern and western regions while eliminating interference from other influencing factors, this paper combines data availability with existing research [60,61], and takes the level of economic development (ED), the level of human capital (HC), the degree of openness to the outside world (OEW), the advanced industrial structure (ASI), the intensity of government intervention (GI), and the level of financial development (FD) as the control variables, in order to eliminate the problem of endogeneity resulting from omitting the important explanatory variables. Among them, the level of economic development (ED) is measured by the logarithm of GDP per capita (lnGDP); the level of human capital (HC) is characterized by the number of college students per 10,000 people (College); the degree of openness to the outside world (FDI) is measured by the share of foreign direct investment (FDI) in the GDP; the advanced industrial structure (IS) is calculated by the ratio of the added value of the tertiary industry to the added value of the secondary industry (T/S); government intervention intensity (GI) is measured by the share of fiscal expenditure in GDP of each province (Fiscal/GDP); and the level of financial development (FD) is taken as the logarithm of the added value of the financial industry (lnFin).

3.2.4. Moderating Variables

Referring to the method of assessing the institutional environment in similar studies [62], the regulation of the institutional environment is measured by counting the number of laws and regulations, the number of policy documents, the number of quality norms, and the strength of implementation that the government promulgates related to data elements. Among them, the number of laws, regulations, and policy documents related to AI technology promulgated by the government can intuitively reflect the degree of importance attached by the government to the application of AI technology in the industrial leapfrog between the east and the west.

3.2.5. Spatial Weight Matrix

Due to the existence of the technology diffusion effect and regional synergy effect, the variables of sustainable innovation and AI technology spillover studied in this paper may have a spatial spillover effect between regions with similar ecological linkage or low-carbon industrial layouts in the east and west, so the spatial weight matrix is introduced. The synergistic nature of the cross-regional flow of green technologies and the intelligent transformation of low-carbon industries is relatively strong, so the spatial weight matrix of green industry–technology synergy is selected, and the inverse of the absolute value of the difference in the density of green technology patents and cooperation between the two regions is used as the weight [63]. This setting is based on the degree of green technology synergy between the eastern and western regions to determine the weights; the greater the difference in green technology cooperation between the two regions, the weaker the spatial synergy, and the smaller the corresponding weights.

$$w_{ij}=\left\{\begin{array}{c}{\displaystyle \frac{1}{\left|G_i-G_j\right|}}\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }i\ne j\\ 1\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }\mathrm{ }i\ne j\end{array}\right.$$

(12)

where $w_{ij}$ is the element in the green industry–technology synergy spatial weight matrix, and $G_i$ and $G_j$ are the green technology patent cooperation densities of the east and west regions i and j, respectively.

3.3. Sample Selection and Data Sources

This paper selects China’s eastern and western provincial administrative regions from 2014 to 2024 as the research sample, which is determined based on the regional division standard of the National Bureau of Statistics of China, covering 10 provinces in the east and 12 provinces in the west. When screening, provinces with missing data on core indicators such as AI technology industry and green innovation are excluded; industry-related data such as manufacturing industry and digital service industry data, which are closely associated with AI technology spillover and sustainable innovation, are retained; and samples in the financial industry and special regulated industries that are weakly associated with the logic of the study are excluded. In order to avoid the interference of extreme values on the regression results, continuous variables such as the intensity of AI technology investment and the number of green patent applications are analyzed according to the 1% and 99% quartiles for shrinking treatment. After screening, 1320 points of provincial panel data are finally obtained.

The data are mainly derived from the following: basic economic data are derived from China’s provincial statistical yearbooks and the China Regional Economic Statistics Yearbook, which are used to obtain indicators such as total industrial output and urbanization level; technology and innovation data are derived from the CSMAR database and patent retrieval system of the State Intellectual Property Office (SIPO), which correspond to the data on the input and output of the AI technology industry and patent applications in each province of China; policy and institutional data are derived from the Chinese Provincial People’s Government’s official website and China Environmental Statistics Yearbook, which are used to extract industrial policy text and environmental regulation-related indicators; and spatial correlation data are obtained by crawling inter-regional technology cooperation projects and industry chain synergy information through Python 3, which assist in constructing the spatial weighting matrix.

4. Empirical Results and Analysis

4.1. Spatial Double-Difference Model Analysis

4.1.1. Spatial Autocorrelation Analysis

In order to verify the spatial distribution characteristics of industrial leapfrog efficiency in the east and west, this paper firstly adopts the global Moran’s I to test its spatial autocorrelation so as to satisfy the application premise of the spatial econometric model [64]. Based on the panel data of 22 provinces and cities in the east and west from 2014 to 2024, the global Moran’s index is calculated by applying the Queen neighbour weight matrix, and the results are shown in Table 3.

Table 3. Global Moran’s index, p-value, and z-value of industrial leapfrog efficiency in east and west, 2014–2024.

Particular Year	Moran’s I	p-Value	z-Value
2014	0.215 ***	0.000	11.265
2015	0.232 ***	0.000	12.083
2016	0.228 ***	0.000	11.956
2017	0.241 ***	0.000	12.531
2018	0.219 ***	0.000	11.427
2019	0.207 ***	0.000	10.892
2020	0.198 ***	0.000	10.376
2021	0.203 ***	0.000	10.641
2022	0.211 ***	0.000	11.058
2023	0.223 ***	0.000	11.724
2024	0.218 ***	0.000	11.389

Note: *** indicate significance at the 1% level.

The results in Table 2 show that the global Moran’s indices of industrial leapfrog efficiency in the east and west are all positive during the period of 2014–2024, all of them pass the 1% significance test, and the Moran’s I value fluctuates between 0.198 and 0.241, which indicates that there is a significant positive autocorrelation in the spatial distribution of industrial leapfrog efficiency. From the dynamic trend, the Moran’s index reaches a peak of 0.241 in 2017, reflecting the strongest synergy effect of industries in the east and west in that period; the index declines slightly after 2020 but still remains high, indicating that there is a sustained spatial agglomeration characteristic of industrial leapfrogging driven by AI technology spillovers and sustainable innovations.

Further plotting of data from 2014, 2019, and 2024 (see Figure 2) reveals that sample points predominantly cluster in the first quadrant (high–high convergence) and third quadrant (low–low convergence), confirming the pronounced spatial differentiation between eastern and western regions characterized by distinct ‘hotspot–coldspot’ patterns. Eastern coastal provinces form high-value clusters with their neighbouring regions, while western provinces exhibit low-value clusters with their adjacent areas. This aligns with the reality of constraints imposed by the ‘technological divide’ and ‘ecological barriers’ between eastern and western regions. The pronounced spatial autocorrelation provides theoretical justification for subsequent construction of a spatial double-difference model, while also indicating that a spatial weighting matrix should be incorporated into the model to capture inter-regional technology spillover effects and industrial synergy effects.

Figure 2. Moran scatterplot of the efficiency of industrial leapfrogging between east and west in 2014, 2019, and 2024.

4.1.2. Identification, Selection, and Testing of Spatial Econometric Models

In order to determine what kind of spatial econometric model should be selected to combine with the double-difference method, this paper is divided into a pre-test and post-test to identify, select, and test the spatial econometric model: the pre-test uses the LM test to determine whether the model needs to introduce the spatial term or not; the post-test is to select the appropriate model, and the specific methods used are the LR test, the Wald test, and the Hausman test, as shown in Table 4 [65].

Table 4. Testing Spatial Econometric Models.

Statistic	Numerical Value	p-Value
LM test (SAR)	102.37 ***	0.0000
Robust LM test (SAR)	75.69 ***	0.0000
LM test (SEM)	138.42 ***	0.0000
Robust LM test (SEM)	104.55 ***	0.0000
LR test (SAR)	31.72 ***	0.0001
LR test (SEM)	27.86 ***	0.0005
Wald test (SAR)	33.08 ***	0.0000
Wald test (SEM)	29.44 ***	0.0002
Hausman test	21.43 ***	0.0009

Note: *** indicate significance at the 1% level.

In the pre-test analysis presented in Table 4, both the LM statistic and robust LM statistic for the spatial autoregressive (SAR) model and spatial error model (SEM) passed the 1% significance test (p < 0.01). The core reason lies in the close technological linkages and factor interactions between industrial leaps in eastern and western regions: Technological breakthroughs within the eastern artificial intelligence industrial clusters propagate to central and western regions through cross-regional collaboration and talent mobility. Conversely, demand signals from central and western industries influence technological iteration in the east [66]. This bidirectional interaction renders industrial leapfrog efficiency markedly spatially dependent, necessitating the inclusion of spatial terms to accurately capture this relationship. From a testing perspective, the LM test rejects the null hypothesis of ‘no spatial lag/error effects,’ indicating that traditional econometric models neglecting spatial factors lead to estimation bias. The robust LM test further eliminates confounding factors such as heteroskedasticity, confirming the authenticity of spatial dependence. This provides direct justification for subsequently introducing the spatial Durbin model (SDM).

In the post hoc tests, both the LR test and Wald test results reject the null hypothesis that ‘SDM degenerates into SAR/SEM’ at the 1% significance level (p < 0.01). This occurs because the spillover effects of artificial intelligence technology encompass the dual pathways of direct and indirect spillovers: The SAR model captures only the spatial lag effects of the dependent variable, while the SEM model captures only the spatial correlation of the error term. The SDM, however, incorporates spatial lag terms for both explanatory and dependent variables simultaneously. This approach reflects both the cross-regional diffusion of AI technology application itself and the spatial linkage of industrial leapfrog efficiency, precisely aligning with the bidirectional transmission characteristics of technology spillovers across both supply and demand sides. The Hausman test statistic of 21.43 *** (p = 0.0009) rejects the random effects hypothesis, supporting the adoption of a fixed effects SDM. This is attributable to the significant non-temporal heterogeneity between eastern and western regions in institutional environments, infrastructure, and other characteristics. The fixed effects model effectively controls for the interference of these individual heterogeneities on estimation results, thereby avoiding endogeneity issues arising from omitted important variables. In summary, this study adopts the spatial Durbin model combined with the difference-in-differences approach (SDM-DID) as the spatial difference-in-differences model.

4.1.3. Spatial Double-Difference Model Regression Analysis

To thoroughly analyze the spatial impact mechanisms of artificial intelligence technology spillovers and sustainable innovation on leapfrog industrial development between eastern and western regions, this section presents regression results for direct, indirect, and total effects based on the spatial Durbin difference-in-differences model (SDM-DID). The direct effect reflects the immediate impact of explanatory variables on industrial leapfrog development within the region itself [67]. The indirect effect measures spillover effects from explanatory variables to neighbouring regions. The total effect represents the sum of both, reflecting the average impact of changes in explanatory variables across all regions. The parameter estimation results for the SDM-DID model are presented in Table 5 below.

Table 5. Parameter estimation results of model (3) and model (4).

Variant	Model (3) (Without Interaction Terms)			Model (4) (With AI × SI Interaction Term)
	Direct effect	Indirect effect	Total effect	Direct effect	Indirect effect	Total effect
AI	0.123 *** (0.025)	0.214 *** (0.041)	0.337 *** (0.056)	0.105 *** (0.027)	0.198 *** (0.043)	0.303 *** (0.059)
SI	0.091 *** (0.022)	0.173 *** (0.036)	0.264 *** (0.048)	0.083 *** (0.024)	0.165 *** (0.038)	0.248 *** (0.052)
AI × SI				0.067 *** (0.015)	0.132 * (0.028)	0.199 *** (0.037)
ED	0.035 * (0.019)	−0.021 (0.034)	0.014 (0.043)	0.042 ** (0.020)	−0.018 (0.036)	0.024 (0.045)
HC	0.057 *** (0.012)	0.098 *** (0.023)	0.155 *** (0.031)	0.061 *** (0.013)	0.105 *** (0.025)	0.166 *** (0.033)
FDI	−0.014 (0.017)	−0.028 * (0.016)	−0.042 ** (0.021)	−0.012 (0.018)	−0.031 * (0.017)	−0.043 ** (0.022)
IS	0.082 *** (0.016)	0.145 *** (0.029)	0.227 *** (0.038)	0.089 *** (0.017)	0.153 *** (0.031)	0.242 *** (0.040)
GI	−0.023 (0.019)	−0.041 * (0.022)	−0.064 ** (0.031)	−0.021 (0.020)	−0.043 * (0.023)	−0.064 ** (0.032)
FD	0.047 ** (0.021)	0.083 *** (0.034)	0.130 *** (0.045)	0.051 ** (0.022)	0.089 *** (0.036)	0.140 *** (0.048)
rho	0.589 *** (0.072)			0.612 *** (0.075)
sigma2_e	0.000 *** (0.000)			0.000 *** (0.000)
N	1320			1320
R2	0.7632			0.7845

Note: Values in parentheses are robust standard errors; ***, **, and * denote significance at the 1%, 5%, and 10% levels, respectively.

Based on the model regression results in Table 5, the various models examining the relationship between artificial intelligence technology spillover effects, sustainable innovation, and industrial leapfrog development exhibit rich characteristics. In model 3 (without interaction terms), the direct effect coefficient for AI technology spillovers reached 0.123. This indicates that, controlling for other variables, a one-unit increase in regional AI technology spillover levels drives a 0.123-unit rise in regional industrial leapfrog development levels. This influence holds at the 1% significance level, demonstrating that the direct driving role of local AI technology in industrial upgrading is both significant and robust. The indirect effect coefficient is 0.214, indicating that a one-unit increase in neighbouring regions’ AI technology spillover levels will elevate the local region’s industrial leapfrog development level by 0.214 units. This coefficient exceeds the direct effect, demonstrating the stronger cross-regional radiation capacity of AI technology, consistent with the spatial decay pattern of technology diffusion. The combined total effect reaches 0.337, demonstrating that AI technology spillovers not only directly enhance regional industrial leapfrog efficiency but also stimulate neighbouring industrial development through technological diffusion, directly validating Hypotheses H1 and H2. The direct effect of sustainable innovation (SI) is 0.091, indicating that a one-unit increase in local sustainable innovation investment raises the region’s industrial leapfrog development level by 0.091 units. This reflects the direct enabling effect of innovative practices such as green technology R&D and low-carbon production models on regional industrial leapfrogging. The indirect effect stands at 0.173, indicating that a one-unit increase in sustainable innovation levels in neighbouring regions elevates the local industrial leapfrog level by 0.173 units. This arises from the strong demonstration effect and shareability of sustainable innovation outcomes, enabling neighbouring regions to achieve secondary transformation of innovations through imitative learning and industrial chain collaboration. The combined total effect of both reaches 0.264, demonstrating that sustainable innovation significantly promotes industrial leapfrogging within the region and its neighbouring areas, thereby validating Hypotheses H3 and H4. Regarding the role of control variables, the direct and indirect effects of human capital level (HC), advanced industrial structure (IS), and financial development level (FD) are both significantly positive, indicating that high-quality talent, optimized industrial structure, and vibrant financial markets enhance industrial transition momentum, whereas the indirect effect of the degree of openness to foreign investment (FDI) is significantly negative. This occurs because the concentration of foreign capital in developed regions such as the eastern coastal areas may create a ‘syphon effect’ on resources in central and western regions, leading to an outflow of local capital, talent, and other factors, thereby inhibiting industrial upgrading.

Further analysis of model 4 reveals that the direct effect of the interaction term AI × SI is 0.067, with an indirect effect of 0.132, yielding a total effect of 0.199. This robustly confirms the significant synergistic relationship between AI technology spillovers and sustainable innovation. AI technology accelerates the transformation of green technological innovations through its data processing and intelligent decision-making capabilities, while the demand for sustainable innovation guides AI technology towards optimization and upgrading in low-carbon domains. This successfully validates Hypothesis H5. Following the introduction of the interaction term, the total effect values for AI and SI become 0.303 and 0.248, respectively, showing a slight decrease compared to model 3. This indicates that the synergistic effect between the two factors is the key pathway for enhancing the efficiency of industrial leapfrog development, rather than the independent action of a single factor. The spatial autocorrelation coefficient ρ increases from 0.589 to 0.612, indicating that incorporating interaction terms enhances the model’s explanatory power regarding cross-regional technology spillovers and industrial synergies. This further validates the interdependent effects of AI and SI at the spatial level, demonstrating that technological coordination between neighbouring regions not only impacts local areas but also influences surrounding regions through spatial transmission mechanisms, thereby improving the overall model’s goodness of fit.

4.1.4. Parallel Trend Test

In order to meet the core assumptions of the spatial double-difference model (SDM-DID), based on the requirement of a quasi-natural experiment, it is necessary to verify whether the experimental group and the control group have similar trends before the policy intervention is imposed in order to satisfy the requirement of constructing the spatial double-difference empirical model. In this paper, we refer to previous research to set the parallel trend test model as follows [68]:

$${EWIL}_{it}=a+{\sum }_{n=-k}^m{\theta }_n{Inst}_{i(t+n)}+{\gamma X}_i+{\mu }_i+{\lambda }_t+{ϵ}_{it}$$

(13)

In model 13, ${Inst}_{i(t+n)}$ is the moderating variable of n periods before and after the institutional environment; if the coefficient ${\theta }_n$ is not significant before the action (n < 0), it indicates that the development trend of the industry in the experimental group and the control group is parallel to satisfy the hypothesis of the model; if the coefficient ${\theta }_n$ is significant after the action (n > 0), the sustained influence of the core factors on the industrial leap can be verified. Setting t = 2019, the period following 2019 represents a crucial policy window during which the CPC Central Committee intensified measures to advance the Western Development Strategy and establish a new development paradigm. This paper employs this boundary to conduct parallel trend tests.

As illustrated by the parallel trend test graph in Figure 3, the regression coefficient following institutional intervention implementation is significantly positive. The parallel trend test validates Hypothesis 7, demonstrating that the new pattern policy for the Western Development initiative has a sustained, progressively intensifying effect on industrial leapfrogging. This outcome is likely attributable to China’s explicit policy direction since 2019, as outlined in the Guiding Opinions on Promoting the Formation of a New Pattern for the Western Development in the New Era, which clearly defined the developmental trajectory and objectives for the western regions. In 2024, the Central Committee of the Communist Party of China reviewed and approved Several Policy Measures to Further Promote the Formation of a New Pattern for the Development of Western Regions, formally launching the new pattern strategy for western development. This has further enhanced the regulatory role of the institutional environment in facilitating industrial leapfrogging between eastern and western regions.

Figure 3. Parallel trend test results.

4.2. Dual Machine Learning Model Analysis

4.2.1. Dual Machine Learning Benchmark Regression

In order to overcome the limitations of traditional econometric models in dealing with high-dimensional covariates and nonlinear relationships, this paper adopts the dual machine learning (DML) model based on the random forest algorithm to estimate the parameters of models 5–7 [69]. The model splits the samples into training and testing sets at a ratio of 1:3 and eliminates the parameter estimation bias through cross-fitting, and the specific estimation results are shown in Table 6.

Table 6. Parameter estimation results for models 5–7.

Variables	Model (5)	Model (6)	Model (7)
AI	0.039 *** (0.009)		0.037 *** (0.010)
SI		0.330 *** (0.033)	0.325 *** (0.034)
AI × SI			0.119 *** (0.026)
Control variables	yes	yes	yes
Time fixed effects	yes	yes	yes
Area fixed effects	yes	yes	yes
N	330	330	330
R2	0.682	0.715	0.736

Note: *** indicate significance at the 1% level.

In model 5, the regression coefficient of AI technology spillover is 0.039, which verifies the direct promotion effect of AI technology spillover on industrial leapfrogging; in model 6, the coefficient of sustainable innovation is 0.330, which indicates that the driving effect of sustainable innovation on the efficiency of industrial leapfrogging in the east and west is significant. The coefficient is 0.119 after the introduction of the AI × SI interaction term in model 7, indicating that there is a synergistic effect between AI technology spillover and sustainable innovation, i.e., AI technology accelerates the transformation of green technology through the data processing capability, while the sustainable innovation demand guides the optimization of AI technology to the low-carbon field, which is in line with the conclusion of the spatial double-difference model, and further verifies the synergistic effect of the two on the industrial leapfrogging in the east and the west.

4.2.2. Robustness Test

In order to ensure the reliability of the research conclusions, the robustness test is carried out from the dimensions of sample adjustment and algorithm optimization to verify the stability of the model and the benchmark regression conclusions, as follows:

(1)

Adjusting the sample scope of the study

This study focuses on regional development interdependencies, with certain Chinese provinces potentially skewing results due to distinctive geographical–economic characteristics or administrative arrangements [70]. On the one hand, from a geographical–economic perspective, Qinghai and Ningxia are situated in the northwestern inland region, possessing limited ecological carrying capacity and population densities significantly below the national average, coupled with low economic activity concentration; Hainan, as an island economy, exhibits weak land-based connectivity with mainland provinces and possesses a relatively small market scale, making it difficult to fully integrate into regional economic cooperation networks. The geographical and economic characteristics of these provinces diverge from the core issues under examination, potentially introducing bias into the estimation results. On the other hand, from an administrative perspective, Wuhan in Hubei, Xi’an in Shaanxi, and Hefei in Anhui, designated as comprehensive innovation reform pilot zones, possess provincial-level economic management authority despite being prefecture-level cities. If these were treated as provincial-level administrative units during variable selection, it would disrupt the uniform analytical framework for provincial administrative units, creating a mismatch between geographical units and administrative tiers and thereby introducing sample selection bias. Therefore, the sample was adjusted in two stages: first, excluding Qinghai, Ningxia, and Hainan due to their distinctive geographical and economic characteristics, and subsequently removing Anhui, Hubei, and Shaanxi where administrative structures introduced sample selection bias, followed by re-testing. The results in Table 7 indicate the following: after excluding Qinghai, Ningxia, and Hainan, the AI technology spillover coefficient is 0.036; after excluding Anhui, Hubei, and Shaanxi, it is 0.054. The core variable significance aligns with the benchmark regression, demonstrating that the inclusion of special regions does not undermine the model’s reliability.

Table 7. Robustness test results.

Variable Type	Adjustment Sample (Exclude Qing, Ning, Qiong)	Adjustment Sample (Exclude Anhui, Hubei, Shaanxi)	Adjust Split Ratio (1:2)	Replace Lasso Regression Algorithm	Replace the Gradient Descent Algorithm
AI	0.036 *** (0.009)	0.054 *** (0.017)	0.041 *** (0.009)	0.009 *** (0.003)	0.027 *** (0.008)
SI	0.135 *** (0.038)	0.222 *** (0.059)	0.114 *** (0.034)	0.088 *** (0.019)	0.093 *** (0.033)
AI × SI	0.024 * (0.014)	0.115 ** (0.051)	0.121 *** (0.035)	0.015 * (0.009)	0.076 *** (0.028)
Control variables	yes	yes	yes	yes	yes
Time fixed effects	yes	yes	yes	yes	yes
Area fixed effects	yes	yes	yes	yes	yes
Sample size (N)	327	327	440	330	330
R2	0.656	0.703	0.712	0.674	0.723

Note: ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

(2)

Adjust the proportion of dual machine learning sample allocation

In machine learning, the split ratio between the training set and the test set is set by human beings, which may introduce bias due to subjective choices. To exclude this effect, the partition ratio is adjusted to 1:2, and the model training and estimation are re-executed [71]. In Table 7, the AI technology spillover coefficient after adjusting the ratio is 0.041, and the parameter estimation results are highly compatible with the baseline regression, indicating that the adjustment of the sample allocation ratio has not changed the core conclusions, and the model is robust to the data partitioning method.

(3)

Replacement of machine learning algorithms

Different machine learning algorithms differ in their ability to capture data patterns due to differences in principles. The baseline regression uses the random forest algorithm, and in order to verify that the conclusion does not rely on a specific algorithm, it is replaced with lasso regression (Lasso) and gradient descent algorithms for re-estimation. Lasso regression achieves feature selection by compressing the coefficients, while gradient descent optimizes the fitting parameters in an iterative manner [72]. Table 7 shows that under lasso regression, the AI technology spillover coefficient is 0.009; in the gradient descent algorithm, the AI technology spillover coefficient is 0.027. The estimation results of the two algorithms are consistent with the conclusion of the random forest algorithm, which proves that the core relationship is not affected by the algorithm selection and the model robustness is strengthened.

4.2.3. Threshold Effect Test

In order to test whether there is a typical threshold value for the driving effect of AI technology spillover on industrial leapfrogging in the east and west, this study analyzes the institutional environment index as a threshold variable [73]. The institutional environment is closely related to the regional technology absorption capacity, and its perfection directly affects the transmission efficiency of AI technology spillover, so the institutional environment index is chosen as the threshold variable.

The threshold (Table 8) effect test results show that the p-value of both single-threshold and double-threshold models is 0.0021, which rejects the original hypothesis, indicating the existence of a double-threshold effect. The estimated value of the first threshold is 1.27, and the estimated value of the second threshold is 2.36. The analysis of 528 sets of data from 22 provinces and municipalities in the east and west from 2014 to 2024 found that only 132 sets of data exceeded the first threshold and 37 sets exceeded the second threshold. The regression results show that when the institutional environment index is lower than 1.27, the coefficient of the impact of AI technology spillover on industrial leapfrogging is 0.312; the coefficient rises to 0.485 after crossing the first threshold, and further increases to 0.673 after reaching the second threshold, which indicates that there is a significant nonlinear moderating effect of the institutional environment on the effect of AI technology spillover. At present, the majority of western provinces have not yet crossed the first threshold, and the space for improvement is significant, as shown in Figure 4.

Table 8. Threshold effect test results.

Original Assumptions	Sum of Squares of the Residuals	Mean Square Error	F-Value	p-Value	10% Threshold	5% Threshold	1% Threshold
Single threshold	0.1632	0.0005	63.27	0.0021	25.6143	30.2156	38.7264
Double threshold	0.1418	0.0004	58.91	0.0021	22.3017	26.8345	35.1028

Figure 4. Double-threshold non-denial domain.

4.2.4. Heterogeneity Test

Considering that geographic location, resource endowment differences, and regional differentiation in the level of AI technology development may make the industrial leap driving effect show different characteristics, therefore, the heterogeneity analysis is carried out from the two dimensions of geographic location and resource dependence, and from the region of AI technology development, to excavate the logic of the differentiation of industrial leap in different regions.

(1)

Geographic location and resource dependence

China’s regional economic development gradient is obvious; the eastern coastal provinces have a deep foundation for industrial digitalization and high-end transformation, the central and western inland provinces have a concentrated layout of resource-based industries, and the transformation is strongly constrained by resource path dependence [74]. Based on geographic partitioning, the sample is divided into eastern provinces and central and western provinces and grouped to explore the heterogeneity of the performance of the drivers of industrial leapfrogging.

(2)

AI technology development region

AI technology is a key force driving industrial leapfrogging in the east and west, but there is a significant gap in the level of AI technology development between regions [75]. Based on the level of AI technology development, the provinces with leading AI technology development are set as “AI technology development advantageous areas”, and the rest are “AI technology development catching-up areas” to test the moderating effect of the AI technology base on industrial migration.

From the regression results in Table 9, on the one hand, in the eastern provinces, the regression coefficients of AI technology spillover and sustainable innovation and the interaction term of the two (AI × SI) are 0.022, 0.214, and 0.089, in turn, which are significantly positive and at a high level, indicating that the synergy between the AI technology spillover and sustainable innovation can efficiently activate the kinetic energy of the industrial upgrading of the eastern part of the country and match the path of high-quality development. In the central and western provinces, the coefficient of sustainable innovation is 0.002 and the coefficient of AI technology spillover is 0.215, but the significance is weaker than that in the east, and the coefficient of the interaction term is 0.133. This is due to the “crowding out effect” of the resource development activities of the central and western provinces on the input of sustainable innovation, and the traditional resource industry occupies a large number of factors, which undermines the development of sustainable innovation; at the same time, the monotonous solidified industrial structure can be used in sustainable innovation development. In addition, the single, solidified industrial structure has great resistance in sustainable innovation, restricting the cultivation of new quality productivity. On the other hand, in the AI technology development advantage zone, the AI technology spillover coefficient of 0.035, the sustainable innovation coefficient of 0.232, and the interaction term coefficient of 0.096 are significantly higher than those in the AI technology development catching-up zone. Advantageous zones realize the deep development of AI technology synergy by virtue of having the perfect AI technology ecosystem and smooth circulation of data elements; catching-up zones are limited by the shortage of AI talents and insufficient technology transformation platforms, and the momentum of industrial leapfrogging is poor.

Table 9. Heterogeneity analysis results.

Variable Type	Eastern Provinces	Central and Western Provinces	AI Technology Development Advantage Zone	AI Technology Development Catch-Up Zone
AI	0.022 ** (0.009)	−0.002 (0.006)	0.035 ** (0.012)	0.008 (0.014)
SI	0.214 *** (0.038)	0.215 * (0.116)	0.232 *** (0.055)	0.186 *** (0.062)
AI × SI	0.089 ** (0.035)	0.133 * (0.075)	0.096 *** (0.026)	0.124 ** (0.074)
Control variables	yes	yes	yes	yes
Time fixed effects	yes	yes	yes	yes
Area fixed effects	yes	yes	yes	yes
Sample size (N)	110	220	110	220
R2	0.735	0.721	0.689	0.675

Note: ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively.

4.2.5. Conduction Mechanism Test

In order to test whether there is a mediating effect of AI technology spillover on the efficiency of regional industrial leapfrogging in the process of sustainable innovation-driven industrial leapfrogging in the east and west, this paper constructs a mediating effect test framework based on the dual machine learning model [76]. The results show that the indirect effect of sustainable innovation on the efficiency of industrial leapfrogging through AI technology spillover is 0.041, with a mediation ratio of 40.39%, Sobel statistic of Z = 4.996, and the Aroian and Goodman statistics are significant (p < 0.01), which verifies that the mediating role of AI technology spillover on the efficiency of sustainable innovation and industrial leapfrogging in the east and the west is significant, as shown in Table 10 and Table 11.

Table 10. Parameter estimation results of the mediating effect of AI technology spillover.

Mediating Effect Pathway	Total Effect Estimate	Direct Effect Estimate	Indirect Effect Estimate	Sobel (Z-Value)	Aroian (Z-Value)	Goodman (Z-Value)	Percentage of Intermediaries
SI → AI → EWIL (total effect)	0.086 ***	0.064 ***	0.041 ***	4.996 ***	4.932 ***	5.001 ***	40.39%

Note: *** indicate significance at the 1% level.

Table 11. Bootstrap test results.

Mediating Effect Pathway	Estimated Indirect Effects	Bootstrap Standard Error	95% Percentile Conf. Interval	95% Bias-Corrected Conf. Interval	Z-Value	p-Value
SI → AI → EWIL (total effect)	0.041 ***	0.008	[0.0211, 0.0539]	[0.0209, 0.0565]	4.08	0.000

Note: *** indicate significance at the 1% level.

5. Conclusions and Implications

5.1. Research Findings

Based on provincial panel data from 1320 enterprises across eastern and western China spanning 2014–2024, this study employs a spatial Durbin double-difference model and a dual machine learning (DML) model to systematically examine the impact mechanisms of artificial intelligence (AI) technology spillovers and sustainable innovation on industrial leapfrog development in eastern and western regions. The key findings are as follows: (1) Artificial intelligence technology spillovers and sustainable innovation significantly enhance industrial leapfrog efficiency in both regions, exhibiting pronounced synergistic effects. Artificial intelligence technology spillovers exert a positive mediating role in the process whereby sustainable innovation promotes industrial leapfrogging. The spatial difference-in-differences model results indicate that the total effect coefficient of artificial intelligence technology spillovers on industrial leapfrog development reaches 0.337, demonstrating a synergistic enabling effect where ‘1 + 1 > 2’. Dual machine learning mediation tests further validate that the mediation rate of AI technology spillovers reaches 40.39%. Specifically, the demand for green technologies generated by sustainable innovation directs targeted spillovers of AI technologies from eastern regions, while these AI technologies in turn accelerate the transformation of green innovation outcomes, forming a synergistic circular mechanism. This conclusion remains robust across sample adjustments and algorithm optimization tests. (2) Heterogeneity analysis reveals significant regional disparities in the driving effects of AI technology spillovers and sustainable innovation. Regression tests by sample grouping indicate the following: from the dimensions of geographical location and resource dependency, eastern provinces exhibit an AI technology spillover coefficient of 0.022, which is significantly positive and relatively high, whereas central and western provinces show a coefficient of −0.002, not only being lower in absolute value than the east but also markedly less significant. From the perspective of AI technological development levels, the spillover coefficient in technologically advanced regions is 0.035, significantly higher than that in catching-up regions. (3) Empirical evidence from eastern and western Chinese samples indicates that AI technology spillovers and sustainable innovation significantly promote leapfrog industrial development across regions, albeit with specific threshold effects. Testing with the institutional environment index as the threshold variable reveals a dual-threshold effect: the first threshold is 1.27 and the second threshold is 2.36. When the institutional environment index is below 1.27, the AI technology spillover driving coefficient is only 0.312; crossing the first threshold raises the coefficient to 0.485, and surpassing the second threshold further increases it to 0.673. Sample analysis reveals that between 2014 and 2024, only 25% of observations crossed the first threshold and 7% crossed the second threshold, indicating that most western provinces have yet to overcome the foundational institutional barrier. These findings theoretically extend the analytical boundaries of risk economics in regional industrial development, revealing the disruptive pathways of risk shocks on technology–innovation synergistic mechanisms. Practically, they provide early warnings for constructing risk-resilient industrial systems during east–west industrial transitions, prioritizing safeguards for technological spillover channels and innovation investment stability. This offers empirical evidence for jointly advancing intelligent and green industrial upgrading across regions, enhancing risk resilience and sustainable development.

5.2. Research Implications

Based on the findings of this paper, China needs to take into account technology spillover and risk prevention and control in deepening regional synergistic development, and form the following policy insights that can be put into practice: First, it should implement a technology-enabled strategic orientation. In the face of risk impacts, AI technology-intensive manufacturing and green technology industries have shown stronger resilience, and social capital should be guided to flow into AI manufacturing, new energy, and other real areas through special funds, supporting green industry subsidies and other policies to strengthen the technological absorption capacity of medium- and high-end industries and the foundation of anti-risk. Second, it should build a supply-side quality and demand-side expansion synergistic mechanism. In view of the dominant role of the supply-side relevance of industrial migration, policy guidance is needed to promote eastern AI technology and green standards to the west of the overflow, synchronized with the improvement of the western digital infrastructure and green energy supply system, to enhance the regional demand side of the technology overflow of the capacity to undertake. Third, it should consolidate regional economic fundamentals and innovation ecology. Taking high-quality development as the core, it should accelerate the linkage and upgrading of the industrial structure of the east and west: the east focuses on the integration of the high-end manufacturing innovation chain, while the west strengthens the green transformation of resource-based industries and jointly promotes the research and development of core technologies, improves the position of the global value chain, and radically enhances the ability to resist risks. Fourth, it should precisely resolve the impact of risks on the factor market. In view of the problems of blocked labour mobility and broken supply of intermediate goods during the risk period, we will establish a mechanism for flexible mobility of AI talents from the east and the west, support enterprises in the west to connect to the intelligent supply chain platform in the east, and guide manufacturers of intermediate goods to improve the level of product diversification and enhance the resilience to risks. Fifth, it should build a chain-wide industrial risk prevention and control system, focusing on monitoring the cross-regional synergistic risks of AI technology spillover-intensive industries and the green industrial chain, establishing an early warning sharing platform for industries in the east and west, and formulating contingency plans to block the conduction and amplification effects of risks in the production network.

5.3. Potential Limitations

Objectively speaking, this study has several limitations: Firstly, the research focuses on the industrial development contexts of eastern and western China, where unique policy orientations and market characteristics exist within China’s industrial transformation process. These contextual particularities may limit the applicability of the findings to other countries or regions. Future research could further examine the relationships between core variables within different institutional environments and industrial development stages. Secondly, this study primarily employs provincial-level panel data for macro-level analysis. While this approach reveals regional patterns, it struggles to capture precise behavioural differences and operational mechanisms at the micro-level. When applying conclusions to enterprise-level practice, calibration and refinement based on specific industry characteristics may be necessary. Subsequent research could select representative enterprises in artificial intelligence application or sustainable innovation, utilizing in-depth case studies and dynamic process tracking to provide more actionable micro-level evidence. Finally, this study’s measurement of institutional environments primarily centres on explicit indicators such as the volume of policy texts and regulatory standards, without sufficiently incorporating implicit institutional factors like regional innovation culture and trust levels in government–enterprise collaboration. These implicit variables may exert latent influences on technology spillovers and innovation synergies. Future research could broaden the dimensions of institutional environment measurement to enhance the study’s comprehensiveness.